计算机技术与发展2025，Vol.35Issue(8)：69-74,6.DOI:10.20165/j.cnki.ISSN1673-629X.2025.0069
基于Vision-xLSTM的遥感图像语义变化检测
Semantic Change Detection of Remote Sensing Images Based on Vision-xLSTM
摘要
Abstract
Semantic change detection(SCD)is an extension of binary change detection.It not only identifies change areas in remote sensing images but also provides detailed semantic category changes,which is particularly important in land cover and land use monitoring tasks.Traditional three-branch convolutional neural network(CNN)architectures and time-consistency-based learning schemes have been widely applied in SCD.However,effectively distinguishing semantic changes and fully modelling temporal dependencies remain challenging.We propose a novel architecture,combining CNN and Vision-xLSTM(ViL),to address the challenges of semantic change detection,referred to as ViLSCD.Firstly,a multi-scale feature enhancement and fusion module is designed to improve the model's ability to represent fine-grained features.Secondly,a differential multi-stage feature interaction distillation module is introduced to enhance the model's perception of change information.Finally,the ViL module is employed to model temporal dependencies fully.Experiments conducted on the Landsat-SCD dataset demonstrate that the ViLSCD model achieves remarkable performance in semantic change detection tasks,with mIoU and SeK scores reaching 90.38%and 64.12%,respectively,surpassing existing methods and confirming the superiority of the proposed architecture in this task.关键词
计算机视觉/遥感图像/语义变化检测/Vision-xLSTM/多尺度结构Key words
computer vision/remote sensing image/semantic change detection/Vision-xLSTM/multi-scale structure分类
信息技术与安全科学引用本文复制引用
张显然,高晗,吴建胜..基于Vision-xLSTM的遥感图像语义变化检测[J].计算机技术与发展,2025,35(8):69-74,6.基金项目
辽宁省高校基本科研项目(LJ222410146057) （LJ222410146057）
